Methods, apparatuses, devices, storage media and program products for determining performance parameters

Abstract

Methods, apparatuses, devices, storage media and program products for determining parameters of a neural network are provided. In one aspect, a computer-implemented method includes: acquiring a first dataset that includes a plurality of face images, obtaining a liveness classification result and a detection result of each of the plurality of face images by inputting the face image into the neural network, and determining performance parameters of the neural network according to a plurality of detection results of the plurality of face images.

1 . A computer-implemented method, comprising:
 acquiring a first dataset that comprises a plurality of face images;   for each of the plurality of face images, obtaining a liveness classification result and a detection result of the face image by inputting the face image into a neural network; and   determining performance parameters of the neural network according to a plurality of detection results of the plurality of face images.   
     
     
         2 . The computer-implemented method according to  claim 1 , wherein the detection result comprises relevant data used to determine whether a face in the face image belongs to a living one or not. 
     
     
         3 . The computer-implemented method according to  claim 1 , wherein the detection result comprises at least one of:
 a face attribute,   a spoof type,   an illumination condition,   an imaging environment,   depth information, or   reflection information.   
     
     
         4 . The computer-implemented method according to  claim 1 , wherein each of the plurality of face images comprises respective annotation information of the face image, and
 wherein determining the performance parameters of the neural network according to the plurality of detection results of the plurality of face images comprises:
 for each of the plurality of face images, obtaining a comparison result corresponding to the face image by comparing the detection result of the face image with the respective annotation information of the face image; and 
 determining the performance parameters of the neural network based on comparison results corresponding to at least a portion of the plurality of face images. 
   
     
     
         5 . The computer-implemented method according to  claim 1 , further comprising:
 evaluating the neural network according to the determined performance parameters to obtain an evaluation result;   based on the evaluation result, acquiring a plurality of training samples related to the evaluation result from a second dataset, wherein the plurality of training samples comprise face images;   for each of the plurality of training samples, obtaining a detection result of the training sample by inputting the training sample into the neural network; and   adjusting weight parameters of the neural network according to a degree of difference between detection results with respect to at least a portion of the training samples and annotation information with respect to the at least a portion of the training samples.   
     
     
         6 . The computer-implemented method according to  claim 5 , wherein the first dataset and the second dataset both comprise real face images and unreal face images, and
 wherein annotation information of a real face image comprises a liveness classification result and face attributes, and annotation information of an unreal face image comprises a liveness classification result and at least one of a spoof type, an illumination condition, or an imaging environment.   
     
     
         7 . The computer-implemented method according to  claim 6 , wherein a number of the real face images in the second dataset is less than a number of the unreal face images in the second dataset. 
     
     
         8 . The computer-implemented method according to  claim 6 , further comprising:
 obtaining the unreal face images through a target acquisition mode.   
     
     
         9 . The computer-implemented method according to  claim 8 , wherein the target acquisition mode comprises at least one of:
 one or more acquisition directions,   one or more bending modes, or   one or more types of one or more acquisition devices used to acquire the unreal face images.   
     
     
         10 . The computer-implemented method according to  claim 9 , wherein at least a portion of the unreal face images comprised in a same dataset are associated with at least one of:
 different acquisition directions,   different bending modes, or   different types of acquisition devices.   
     
     
         11 . An electronic device, comprising:
 at least one processor; and   one or more memories coupled to the at least one processor and storing programming instructions for execution by the at least one processor to perform operations comprising:
 acquiring a first dataset that comprises a plurality of face images; 
 for each of the plurality of face images, obtaining a liveness classification result and a detection result of the face image by inputting the face image into a neural network; and 
 determining performance parameters of the neural network according to a plurality of detection results of the plurality of face images. 
   
     
     
         12 . The electronic device according to  claim 11 , wherein the detection result comprises relevant data used to determine whether a face in the face image belongs to a living one or not. 
     
     
         13 . The electronic device according to  claim 11 , wherein the detection result comprises at least one of:
 a face attribute,   a spoof type,   an illumination condition,   an imaging environment,   depth information, or   reflection information.   
     
     
         14 . The electronic device according to  claim 11 , wherein each of the plurality of face images comprise respective annotation information, and
 wherein determining the performance parameters of the neural network according to the plurality of detection results of the plurality of face images comprises:
 for each of the plurality of face images, obtaining a comparison result corresponding to the face image by comparing the detection result of the face image with the respective annotation information of the face image; and 
 determining the performance parameters of the neural network based on comparison results corresponding to at least a portion of the plurality of face images. 
   
     
     
         15 . The electronic device according to  claim 11 , wherein the operations further comprise:
 evaluating the neural network according to the determined performance parameters to obtain an evaluation result;   based on the evaluation result, acquiring a plurality of training samples related to the evaluation result from a second dataset, wherein the plurality of training samples comprise face images;   for each of the plurality of training samples, obtaining a detection result of the training sample by inputting the training sample into the neural network; and   adjusting weight parameters of the neural network according to a degree of difference between detection results with respect to at least a portion of the training samples and annotation information with respect to the at least a portion of the training samples.   
     
     
         16 . The electronic device according to  claim 15 , wherein the first dataset and the second dataset both comprise real face images and unreal face images, and
 wherein annotation information of a real face image comprises a liveness classification result and face attributes, and annotation information of an unreal face image comprises a liveness classification result and at least one of: a spoof type, an illumination condition or an imaging environment.   
     
     
         17 . The electronic device according to  claim 16 , wherein a number of the real face images in the second dataset is less than a number of the unreal face images in the second dataset. 
     
     
         18 . The electronic device according to  claim 16 , wherein the operations further comprise:
 obtaining the unreal face images through a target acquisition mode, and
 wherein the target acquisition mode comprises at least one of: one or more acquisition directions, one or more bending modes, or one or more types of one or more acquisition devices used to acquire the unreal face images. 
   
     
     
         19 . The electronic device according to  claim 18 , wherein at least a portion of the unreal face images are associated with at least one of different acquisition directions, different bending modes, or different types of acquisition devices. 
     
     
         20 . A non-transitory computer-readable storage medium coupled to at least one processor having machine-executable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
 acquiring a first dataset that comprises a plurality of face images;   for each of the plurality of face images, obtaining a liveness classification result and a detection result of the face image by inputting the face image into a neural network; and   determining performance parameters of the neural network according to a plurality of detection results of the plurality of face images.